My friend Mike pointed out this morning that tech people pushing the simulation idea are just rationalizing God. They don’t want to believe in God, so they believe in the simulation instead. Solid take. I’m not here to argue with it. I just want to lay out why the simulation thing actually makes sense to me.

Unlike God, the simulation hypothesis is built on pure logic, not belief.

Here’s the logic: if I can put you in a simulation, how do you know nobody put me in one? The whole argument only falls apart if we can’t actually build a good simulation. So let’s look at what we can already do today.

Computers are writing code that writes code. They’re creating full games, understanding video, generating video, understanding audio, generating audio. They can spin up a living, breathing video game that builds itself as you play.

We also have Neuralink-style tech that reads your brain signals and writes signals back in. People already control computers with pure thought. If your brain can talk to the machine, the machine can talk straight back to your brain.

At the same time, companies are training AI models to give robots real senses—smell, touch, the whole package. Right now a robot grabbing a cup has zero idea what glass, ceramic, or metal actually feels like. It follows rules (“be gentle with glass”), but it doesn’t feel it. We’re actively teaching machines the exact sensation of slippery glass, soft velvet, fresh-cut grass, the smell of roses. If we can teach robots, we can teach computers. And if we can teach computers, we can code an entire believable world—sight, touch, smell, gravity—and pipe it straight into a human brain without the body ever moving.

You could literally be floating in a Matrix pod right now. The computer is telling you you’re staring at a keyboard, feeling its keys under your fingers. You’re holding a mouse, sensing its shape and weight—all of it beamed directly into your brain, bypassing your real body completely.

If we’re about to pull this off (and we are), how do you know it hasn’t already been done to you?

Cheers,
Kirill

If you liked this, you might also like to read my take on forecasting memories based on memories from other people -

If that was of interest, then you’d also want to read about self driving, and why the age of AI is going to radically change how you think about technology -

Lastly, I am about to release a new version of Postal. If you are a Mac user, grab the current version, and we will automatically upgrade you to the next when it’s ready. It’s going to be VERY different, but also super AI. It’s rad. I Promise.

What if memories are not unique, and instead they could be predicted, just like the next token?

It’s a weird question because most of us would like to believe that we are special and therefore our memories are special and unique too. But is that really the case?

Currently, AI chatbots have the ability to create memories about individual users. They store those memories in context and when you make a request for new information, the AIs try to understand if any of existing memories about you are relevant to this conversation. It’s very basic, and it is not infinitely scalable because the more memories they store, the harder it is to give a great response. The AIs work best with the clearest information, and the more you pollute it with other things, the less sure it gets.

What if we could share memories? I don’t mean just between people close to you, but to share memories around the world?

If I’m a dad who lives in Austin and has a dog, then my memories about Austin will also overlap with other families that live in Austin and have a dog. We already share memories in a way, when we do a Google Search. Say I want to find a few fun things to do with the family on a road trip to Houston. I’d find a dozen things to do on the internet, but I will also discard things like bars and nightclubs and prioritize things like the Space Center and the Nature Museum. I’d do that that converge general results around my family’s interests.

Why don’t we just let AI use this information in aggregate?

We could train AI on this shared context and enable it to forecast memories. If my family and I went to the Butterfly Museum and walked from there to the Japanese gardens, there is a chance that 10,000 other families like us did exactly the same thing. Therefore, this is a useful memory to other families like us.

If a car drove over a pothole at a particular GPS coordinate, chances are that thousands of other cars hit that same pothole. We should share that information.

The way we think of memories is something private, something that we don’t want to share though, but we could anonymize this knowledge and train a transformer on what those memories are and how to use them. Your personal data will remain personal, but the transformer will now know how to predict the next token of memory. The best part, this can be done by leverage existing architecture.

By sharing memories, we could literally give AI an ability to see the future.
Your future is just somebody’s past.

-Kirill Zubovsky

p.s. This also answers the question of what happens when LLMs have scraped and trained on all of the content available on the internet. If there’s no way to make money from making content, there won’t be any more training data. Well, your own experience with the machine can now become the data for the machine. From there, it never ends.

p.p.s. So, what are the chances that we are already in the simulation, providing endless data to the machines?

My wife isn’t quite so comfortable yet. She’ll engage FSD, but turn it off when she feels like she could make a better decision. This is not the way of the future, but to understand that, you first have to understand FSD, and the human nature.

AI is about to completely upend every industry it can touch, and it is going to touch every industry. Let me use Tesla as an example to show you what the future holds, and how to make the most of it, if you don’t want to be left in the dust.

How FSD Works

Somewhere around 36-60 times a second (depending on the camera), your car looks at the road around you through eight different cameras. It then converts that information into data and sends it to a neural network. It sees 360 degrees, far beyond human peripheral vision, without fatigue or distraction. The car looks forward, backwards and sideways, it looks near you and up to its maximum visual range. It makes a decision on where to go, continuously.

Can you pay attention to the road, dozens of times each second, for the duration of your entire trip? Can you look behind you, next to you, and ahead of, process all the information, and make a split second decision, and then execute it?

That thought alone should make you understand that a computer is a much better driver than you are because it is able to look everywhere all at the same time, continuously without getting tired, without getting distracted, without getting annoyed. It doesn’t eat, it doesn’t talk on the phone, it does not deal with the kids in the back seat. The computer just does one job, and it does it perfectly.

Here’s another fun fact (I think?). The car does not actually think about the objects on the road. It doesn’t know what a pedestrian is, or what a stop sign is, what a traffic light is. It doesn’t concern itself with “an aggressive redneck” or a “slow grandma.” The computer has 10 years of training data to know how to behave around those objects, and it just quietly responds to the data. What you see on the screen, a visualization of the road, the people, the signs, that’s just a parallel process which displays that data for humans; the car doesn’t care.

You could say that FSD isn’t flawless yet, and it’s true, there are very minor edge cases. AI is still learning from billions of real-world miles. In Austin and in San Francisco though Tesla Robocab is able to operate fully autonomously, so we know the technology is nearly there. In case of Robotaxi, a remote employee could help out in case of a real challenge, but in case of FSD in my car, the worst thing that could happen is my car would ask me to take control. The difference is, in my car I could spend 99.99999% of the time without really engaging in driving, while non-FSD drivers spend 100% of their time driving.

Why are the humans so afraid of AI?

So why is it slightly freaky for my wife to engage in self driving, if the self driving is so unbelievably good? It’s the human nature!

When we drive ourselves, with hands and feet, we do certain things that we’ve learned to do in order to make our driving better in relation to other human. When you see a red light ahead you, you might slow down. When you see a truck merging in the lane next to you, you might give it space. When you see a car slightly swerving in the lane next to you, you might pull away just in case…etc. We do many little things like this because we’ve learned it helps us drive safer, but the car does not care.

The self-driving car just wants to know if it can make the right move in this moment of time. It doesn’t need to move away from another car, if it does not need to avoid it, but if thinks that it needs to avoid it, it moves over quickly!

The coolest part? Tesla AI didn’t learn any of these things because someone sat down to teach it, to explain what was right and what was wrong. AI just watched years of driving data, and learned to follow along.

You see where I am going here?

AI Takeover

We are at the very-very beginning of what is possible. The AIs are going to take over every single industry. As long as there’s data on what is right and what is wrong, that AI will figure it out, and it will get better and better at doing those things, until it’s doing them perfectly.

This isn’t just about driving. AI is already transforming:

• Healthcare: Models like those from Google DeepMind analyze scans faster than radiologists, spotting cancers with superhuman accuracy from vast image datasets.

• Finance: Algorithms trade stocks in microseconds, learning from market history to predict trends better than any human analyst.

• Creative Fields: Tools like Midjourney generate art from text prompts, trained on billions of images, challenging artists to collaborate rather than compete.

• Code: Claude Code and tools like it are making it possible to create in days what would previously take many months.

In each case, AI excels where there’s abundant data to learn from.

You can either fight this idea and pretend that it’s not going to happen, or embrace it, and plan for the vibrant future where mostly everything that used to be done by humans is now done by AI.

To embrace AI , you must let go of ego.

Letting go of ego means recognizing AI as a tool, not a threat. For my wife, it’s unlearning those micro-adjustments we’ve honed over years of driving. In the workforce, it’s similar. Great coders, for example, might fear AI writing code, but the winners will use it to amplify their ideas. Resistance often stems from job fears or loss of control, but overcoming it starts with using AI in your own life.

Audit your day, or your job. What tasks are data-driven and repetitive? Delegate them to AI and focus on creative, human elements. Try FSD if you have a Tesla, or explore free AI tools like Grok for problem-solving.

It’s really simple to be afraid, and to resist, and to worry. Making yourself useful in the AI future will require hard work, just like before, but now that work can be amplified many times over.

We’ve all go an enormous power on our hands. Use it.

-Kirill

p.s. This post is brought to you by Postal, a todo app nobody’s dream off. 🙀

Every time I read one of those hot takes, I quietly look at my GitHub.

How could all that negativity be true? I have been able to create dozens of apps this year; some are long gone, but some have made it to production, and work incredibly well. How can I see SO MUCH GOOD, where others see only errors?

I could not solve this puzzle until a few days ago when I realized — my mental model was entirely wrong. What I think “vibe coding” means is not what most of you probably think.

My code doesn’t get worse because you call it by a different name, but what you think about it changes dramatically based on the name. That’s not very good, and I should fix it.

The Two Definitions of “Vibe Coding”

Definition A (what most critics mean):

Sit on a couch → open Lovable/v0/Bolt → type “make me a Twitter clone with auth and payments” → sip coffee → pray it works.

Zero planning. Zero architecture. Zero understanding. Pure vibes. Let the AI drive.

This is what people picture when they dunk on “vibe coding,” and you know what, they are right to dunk on it. That style breaks the moment you leave the happy path. It produces beautiful corpses—apps that look perfect until someone actually uses them.

Definition B (what I’ve been doing all year):

AI-assisted development on steroids.

I am not asking my AI for miracles. Instead, I design the architecture, plan the details, research unknowns, write the hard parts, and at the end, I still review [some of] the code that goes to production.

The difference is that 60–80% of the boilerplate, CRUD, UI scaffolding, refactors, tests, and even tricky algorithm implementations are done by AI first. I use AI to plan what to do, I use it for search, use to figure out how to do the product, but then, when it’s all ready, the AI writes first draft. It can do it faster and often better than I would have on my own. From there, we follow the same process to iterate and to finish till it works.

I’m not “vibing.” I’m directing an orchestra!

Taste and Experience Matter (for now)

The people who fail with “vibe coding” treat the AI like a junior dev, they push the AI around, set unrealistic expectations, and fire it for asking questions. The people who succeed, treat it like the world’s fastest, most obedient senior dev who never sleeps and will happily rewrite the same component 27 times until it’s perfect.

The difference isn’t the tools. It’s architectural taste, system design intuition, and ruthless code review—exactly the same skills that separated great engineers from average ones before AI existed. LLMs didn’t remove the need for skill.

Can We Please Retire the Phrase?

“Vibe coding” is now a polluted term. It means two completely opposite things depending on who’s saying it.

Right now, the loudest critics are reviewing a tricycle and declaring that all electric vehicles are toys—while a growing number of us are quietly shipping production apps at 10x speed and wondering what the fuss is about.

The tools aren’t holding us back anymore.
The words might be.

— Kirill Zubovsky

Since you’ve made it this far, a few more things to read:

A few weeks ago I bought an NVIDIA GDX Spark so I could learn how to run AI models locally. In the first few days I was able to run Llama 4 and OpenAI OSS, achieving pretty good results at a low cost.

While entirely not conducive to building a billion dollar company, this weekend I decided to repurpose my Spark for something else — a Minecraft server.

I’m not a server guy, but I have a general idea to how things work. It took about four hours of trial and error and talking to my AI in order to write the code, test in and iterate to where my family is now able to play Minecraft without it crashing.

I used local gpt-oss:120b running on my Spark to write a lot of the code, but I also used Grok 4.1 for efficiency.

The most annoying part of local inference is poor access to up-to-date information. When something goes wrong (like a version mismatch on some Python package), local AI relies on its memory and logic, while Grok can quickly query all of the internet and get just the right answer. This will of course improve over time.

That’s it, there is no grand reveal this time. I just wanted to tell you that if your kids want to multiple Minecraft on a local LAN, and they want to use both a Mac laptop and a Nintendo Switch, you can now enable them to do so.

If you want to setup your Spark to run Minecraft, all the details are on Github. You should be able to pull the repository and one-click install and be ready to play.
👉 https://github.com/kirillzubovsky/nvidia-spark-minecraft-server

That’s it for today. Ask me questions if you have any.

Have an epic week ahead!

-Kirill.

If you want to follow along, you can:
- YouTube (playlist)
- X Nvidia GDX Spark (community)

AI people love to invent words to describe otherwise pretty mundane things, and the word “Agent” was just another one of them. Nobody knows who came up with it, but at one point in the past, code blocks got renamed to agents almost overnight, and the agentic revolution was born.

Agents are code that are constrained to a particular area of expertise. Most importantly, they are supposed to be able to work autonomously within that area.

Remember The Matrix?

Instead of being busy doing everything in the world, Mr. Smith had only one goal - to maintain order and security within the matrix, and to eliminate the threats. Nobody told Smith what to do, he just did what he thought was right. In the same way AI agents are constrained to doing only a specific subset of tasks, and doing it on their own, really well.

General LLMs are just dumb text generators. They know everything about everything, but they aren’t particular good at any one task. If you want a refresher on the origins of LLMs, it’s here—

Agents, on the other hand, are code loops that use the underlying LLMs. One agent “knows” how to book tickets, another one that can update your calendar, another one writes code, and yet another adds milk to your refrigerator when it runs out.

This is helpful because it makes the agent more focused, yielding better responses to specific questions. Being a focused system, an agent gets to prioritize responses for a particular area. Better results for you, faster computer time for the underlying models. Win-win-dial-tone.

These agents work in three easy steps:

1. First. Sense the environment. This is where they read your input, and analyze it to, trying to understand what you actually want.

2. Second. Make a decision about the next steps. Using either predefined steps, or some kind of internal logic on how to behave based on the inputs, the agent pick the next steps, be that to search the internet, to parse your data, call other applications, or anything else like that.

3. Third. Act on the input and repeat. At this stage, the agent does what it planned in step 2, collects all the data, analyzes it, and then sends it back to step 1 to repeat until the original problem was solved.

So that’s what it is, in short LLM agents turn passive language models into proactive goal-seeking loops that “think”, use tools, and iterate until the job is done.

That’s it, no magic, no moat, just code that is told to do the same thing over and over and over again.

Next time someone talks to you about AI Agents with grandeur and valor, you will know what they are talking about. It’s a bicycle, sold a 747.

— Kirill.

Raf and I saw a problem, solved a problem, and continued onwards trying to figure out what consumer application would benefit the most from the solution to the problem. The guys who did mem0 looked at the same problem, and turned it into “open source memory for AI agents.”

I don’t even know what that sentence means. The technology layer they are providing should be implementable by someone with a half a brain in a weekend. But apparently, that’s enough to start a company and to raise tens of millions of dollars these days.

While we looked at a problem and continued looking up to the next problem that emerges from it, someone look down and realized that thousands of people will run into the same problem, and either won’t be able to solve it, or would prefer an existing solution.

This had literally *never* occurred to us.

Look, I am not gloating about being smart here. Although Raf was pretty much top of his class in Engineering Science, the hardest program at UofT. I am the dummy one in this group. But, my point is quite the opposite — our ability to understand the thing got in the way, big time.

I might not be the right person to do this company. Reselling the same 3 functions while pretending it’s the holy grail of the AI future isn’t exactly my cupt of tea. But with that in mind, this is a note to all my engineering-minded readers, you really don’t have to think too hard.

In fact, just stop thinking.

Just because you understand something, there’s a very, very high chance that most people around you don’t. Do with that what you will.

-Kirill.

p.s. If you want to read the post from last year —

As a child you may not realize it, but life is hard, and your parents are always afraid. They see somebody else have means, but they don’t see a way by which they too can have things.

An illusion is formed that the rich have somehow got to where they got by some shady means. There is envy, maybe resentment towards the wealthier folks. The path to success isn’t viewed through hard work, creativity, connections and cleverness and some luck, no, it is seeing as something impossible, out of reach, out of consideration.

That’s certainly how I grew up. No one in my family ever hated on the rich, but in hindsight, I can tell there was a layer of largely self-imposed believe that the rich were unreachable. Sadly, this was a self-fulfilling prophecy.

It took a long while to understand this, and I am still unlearning the bad habits that come as a result of this thinking. There’s no alcoholics anonymous for learning to think clearly, for figuring out how to unchain thyself, for remodeling your brain to take on the impossible.

It’s still a work in progress for me.

Sadly, while I am still working on making the switch, I see others fall right into the same trap. So many folks think they are stuck in their space, feeling entitled but unable to get what they want, channeling their energy into hate rather than forward momentum.

You can see this quite well in the regular folks that hate Elon Musk. They hate him for being a billionaire, for meddling into politics, for having the will, the strength, and the power to see things through. It’s not hate based on facts and logic, it’s entirely based on feelings, on their inability to see how they too could one day make a difference. They think they make a difference by shouting, producing nothing but noise.

You can look at Elon and see a billionaire who has more than you, or you can see a guy that made driverless cars, autonomous rockets, everyday robots, who saved free speech, pioneered a brain interface to computers, and in the process enabled hundreds of thousands of people to live better lives.

The situation that has formed your reality to date may have been entirely out of your control.

How you see the world around you, and how you envision your future, that’s is always entirely up to you.

It’s 2025, and we can watch satellites being launch, from the point of view of the satellite. How amazing is that?!

Per Aspera Ad Astra. Through hardship to the stars.

If for whatever reason during the install your spark SHH service experienced an error, like it did in my case. You will not be able to connect to the device until you connected to a monitor, keyboard and mouse, and login into the device.

The simplest way that I found was to go to Best Buy and get a $10 keyboard and a $10 mouse with a USB-C dongle and then connect that to Spark. If you don't have a compatible screen, use any basic TV with an HDMI cable. If you have to, BestBuy sells monitors for like $200; worth a visit.

Once you have connected these peripherals and you got into the Spark, here's what she do:

1. Regenerate the Missing Host Keys

Run this command to automatically generate all standard SSH host keys (RSA, ECDSA, Ed25519, etc.) in `/etc/ssh/`:

sudo ssh-keygen -A

This is safe and non-destructive; it only creates keys that don’t exist. If prompted, accept the defaults (no passphrase needed for host keys).

2. Fix Permissions on SSH Files (If Needed)

Ensure the keys and config have correct ownership and permissions (owned by root, readable but not writable by others):

sudo chown -R root:root /etc/ssh
sudo chmod 600 /etc/ssh/ssh_host_*_key
sudo chmod 644 /etc/ssh/ssh_host_*_key.pub
sudo chmod 644 /etc/ssh/sshd_config

3. Restart the SSH Service

This will just make sure the service starts with new configurations

   sudo systemctl restart ssh

4. Verify the Fix

Check the service status (should now show “active (running)” without errors)

 sudo systemctl status ssh

Confirm the port is listening:

sudo ss -tuln | grep :22

After this, my Spark became instantly available on the network and I was able to use NVIDIA Sync application to connect with it. You can also ssh directly into the spark.

ssh your_spark_username@your_spark_local_network_ip

That’s it, it’s that simple. Unfortunately it was not as simple without a direct connection to the Spark system to see what was going on inside.

Here’s me explaining why the initial Spark setup is rather user-unfliendly:

— Kirill

p.s. If you like this type of content, please join me on X Nvidia Spark Group

Before we continue, if you want to watch the video, please follow on:
- YouTube (playlist)
- X Nvidia GDX Spark (community)

It’s a great question and I bet it is something many people are asking themselves right now. I totally get it—it’s a smart, cost-effective approach to use the APIs, and it doesn’t require tinkering with local hardware. But honestly, that’s not why I’m diving into this.

For the last few months, while building all the tools I’ve been talking about, I’ve been burning through about $1,000 a month on Cloud Code alone. I am not waiting for no limits, I just go straight for the API and get the Opus model to go full speed.

It sound expensive, but the price isn’t the point. Even if you’re looking at $12,000 a year for top-tier hardware, that’s virtually free compared to what big tech companies pay their engineers for similar work. Salaries in this space are massive, so hardware costs? Practically negligible. I am blowing through 1 Spark every 3 months only; I could do 3-5x if you let me!

The real driver to try a local device like Nvidia Spark here is the future of AI—and how we use it.

I see a world where we seamlessly switch between online and offline models. Offline AI is going to handle the mundane, working 24/7, augmenting our data, streamlining our workflows, and just... working while we sleep, and then working while we are working, and then working when we are not working. For the compute-intensive stuff, we’ll offload that work to the cloud GPUs, or more specifically, our local AIs will request help from the cloud AIs.

Local for everyday efficiency, cloud for the big guns.

Personally, I don’t want to be at the mercy of what OpenAI (or any other giant) decides tomorrow. Will they secure more funding? Will they run into data center restrictions or shifting regulations across countries? I love AI—it’s my daily go-to tool—and I would rather not lose access because of external factors.

Losing access to AI at this point will be equivalent to losing clean water.

Looking 18 months ahead, local AI is going to explode. I want to be ahead of the curve, fully immersed when it becomes the norm, not scrambling to catch up.

NVIDIA Spark isn’t just hardware—it’s a gateway. It does everything a standard NVIDIA chip can, but right on your desktop. Once we master local setups, scaling to NVIDIA’s cloud for complex workflows is seamless. It’s all in one ecosystem, making experimentation easy and powerful.

Is it the absolute best local AI hardware out there right now? Maybe not. But it’s NVIDIA, and that means compatibility, reliability, and a playground for innovation. I want to tinker, test limits, and see what’s possible (and what’s not).

Once dialed in, this setup runs 24/7 for way less than constant cloud reliance. It’s efficient, it’s yours, and it’s future-proof.

— Kirill

p.s. If you find this interesting, follow along on X, YouTube, or here on Substack.

A small model (Gemma 3, for example) is like a Casio watch - light and cheap, and tells time, maybe has a calculator. Meanwhile, a large model (like LLAMA 4 from Meta) is like a mainframe from the 80s, it takes the whole room.

Which ones should we use, and can we actually use them?

With great power, comes great power consumption!

It’s a wild west in the LLM space right now. We are all racing forward, but not sure we know where to. We are just racing, hoping that along the way someone will discover something, and it will change everythign.

Are we going to get one super model that can do everything and is only accessible to the super rich? Are we going to stop with AI, and go back to doing things by hand? Are we going to put giant solar arrays into space and crunch numbers in floating containers, cooled by dark matter?

Well, the latter actually might be the case, but …

Libraries, I think the future will be full of libraries.

We had libraries in the past, and while they were tedious, they also worked quite well for some use cases. You could go to a Library and get any book that you wanted, on any topic that you could find. When finished, you could go get another book, from that, or another library.

Library is an infinite source of knowledge, capped only by time.

There’s a very high chance that just like we used to have books, we will now have libraries full of models. In fact, using actual libraries as data centers might be an interesting concept. You go there, and you get to have access to powerful models, but only while you are physically tethered to the fiber cable. Perhaps directly through a Neuralink.

These libraries of the future will be “books”, but instead of physical paper, you will be connecting to different models. Some models will be good for coding, others for cooking, for history and for mathematics…etc. In fact, there will be libraries dedicated to those subjects, with books on different sub-topics.

In the future, you will connect to a provider that serves physics models, for example, and then you will choose whoever is the best in nuclear physics, and chemistry.

Supervising it all there will be a librarian, if you well, who will be able to reroute your information request to just the right place. You won’t have to talk to her, it will be just an AI with a specific set of reasoning skills, quickly guiding their audience to the right data.

Some of these ebooks you will take home for free, some of them, you will access remotely for a reasonable price, based on time and value.

I think this will solve a lot of problems with these below the bottles that are too large to effectively run on laptop. Furthermore, just like most people don’t read books on ALL the topics, most people will not need all the models. There will be cheap publicly available models serving AI slop, news, and some basic conversations. Paid models will do that much, much better. Other models you’ll be able to buy, or check out from those libraries and bring home. Math and chemistry libraries for the nerds, gaming libraries for others.

— Kirill.

Addendum:

Why does this even matter? Can’t we have local models and be happy with them?

Existing AIs have a few constrains. Models like Grok or Claude Code that run via the web API (aka, from their servers to my home computer) are very-very good. They however require an internet connection, and they do cost quite a bit. Models that can run locally on my MacBook Pro, thus only constrained by cost of a powerful laptop, those models have issues:

What issues, you ask?

1. Local models are capped in their training data at the time of release, which means if I ask a question about something that happened this morning, the will not have a clue. These models need to be updated with new content if you want them to access to that content, and while there are multiple ways of doing so, it doesn't work out of the box, it require rather a lot of effort.

2. On top of that, the small local models are only great a few things, not everything. A programming model might do well at programming, but it won’t be able to do project management, or creative writing. This requires switching models around, and who wants to do that?

3. And last, but not least is reasoning. A good model is no longer just a predictive text generator, it knows how to think. Not in the way you and I would think quite yet, but it’s getting there. It know how to figure out what I actually want to know, and give me that precise bit as the answer.

So yeah, we need more than just local models. A lot more, and for a lot less, if we want to truly make this technology the revolution that it wants to be. This has to be the new internet, available for all, with added benefit to those with money.

The guys who bought eventually shut it down. Two years later, the biggest players in domains still haven’t implemented a tool even 1/2 as good. It was like building a space rocket, and then selling it to ants.

After that, I swore that I will not do another thing that I didn’t care about. Time is the only scarce resource; it’s critically important to invest it wisely. Fast-forward to now, and here we go again, I am working on an SEO analytics product.

Do I care about SEO? NOPE! Every day is a drag.

Nothing against SEO or the people. It’s a necessary evil, and the people are doing their best work. It’s the same as I felt in the domain space. It’s a space, it’s a job, it’s just not something I care to wake up and think about for 20 hours a day.

Plenty of people would find it totally fine to wake up and just do something;
not me.

I mean, like I said in my previous post, I do like our product, and I find the output very, very good! The whole point of us creating EasySEO (previously named Ga4hell) was to automate a process that we didn’t particularly wanted to do manually. It works, and it works amazing well!

Here’s an example of our dashboard for AiTinkerers.

You see, it looks pretty good. A TON of valuable suggestions. You can just feed the export to Claude Code and watch it transform your website. It’s so awesome, most people are blown away by the possibilities. And yet…

We made the tool to get away from doing a process, but to sell the tool to others now requires boiling in that process every damn day.

The prize for chewing glass is more glass

Is this normal? Does Elon still enjoy making cars, does Gary V still enjoy making videos, does anyone actually care about the thing they do, or is the steady-state of anything is simply vanilla?

Actually, I think the latter is true, and the only way to escape it is to find the space hard enough to keep pushing. That’s why the best minds right now are pushing AI and Robotics, because you can.

Do hard things. Everything is a drag.

Writing is thinking, and thinking is learning.

There is a part of me that wants to believe that SEO is going to be more popular than ever with the emergence of AI tools, but there is also a bigger part of me that starts to believe that SEO is gonna be gone in five years. I mean like completely gone, in the same way as dial up Internet is gone-gone.

The only part of me that wants to believe in SEO exists because we built a tool for it (read previous post: https://novice.media/p/should-we-sell-gains-over-pains)

Is there a world where SEO still lives?

I see a future where users get instant, synthesized answers from AI without ever needing to search or click through to a site. ChatGPT and such are currently providing citations to answers, but let’s face it, how many people follow through to see the source? Google, while showing you the results, is also summarizing the best content at the top. No one is doing the research anymore, we just want it done and packaged in a single answer.

Result? Just as dial-up couldn’t compete with always-on connections, SEO can’t survive in a landscape where the “search” happens invisibly inside the AI.

Heck, it’s not just search, based on the recent demos, the whole internet is going to happen inside of these AI applications.

Some say that “Answer Engine Optimization” is the future, but I am starting to wonder if that is the future, or just our hope for survival? If you’re not cited, you don’t exist, but do you exist IF you ARE cited?

In the 90s, we romanticized the screeching modem as “the future,” but once fiber optics hit, no one looked back. Some countries skipped fiber entirely and went straight for cellular internet everywhere. SEO’s seems to be a clunky intermediary born for a pre-AI era of human-curated links. AI isn’t just faster dial-up, it’s Starlink, Internet delivered with no wires at all.

It is possible to assume that, as the future AI takes over, everything becomes so vanilla and so boring that human-generated content once again becomes the focus of our desires, and we seek for ways to resurface it. But can one be on it? Will this happen in 5 years, or will it take 25?

Maybe that's what it is. Maybe the best one can do is to squeeze the maximum value out of their product this very minute, assuming that the future is not guaranteed. If the future comes fast and it’s favorable, great, but if it doesn’t, then at the very least we find out and adopt fast.

— Kirill.

p.s. You might also find it valuable to know what Reddit think about SEO in 2025.

What we have is a tool that can significantly improve your website by recommending changes that add value to your users' experience. As a result, the users stick around and buy your product. What we have been selling so far was a tool to improve AI and search rankings, thereby making your website more relevant to potential users.

Sounds different, but it's actually two sides of the same coin.

The problem I had was this — To sell the SEO product, I have to sell to someone who buys into the value of SEO and is already committed to seeing that value grow. To sell to the end consumer, I need to sell to someone who’s got a dollar and wants to buy a can of soda, RIGHT NOW.

You know that feeling when you know that vegetables are good for you, but right now you want steak and ice cream? Same thing.

I think we’ve been selling this tool wrong this whole time!

I know this because I used our own tools on my own websites and it’s been great, but I wouldn’t be able to sell this long-term vision to my own self, only the instant gratification.

Breckworks.com for example, I implement 40+ changes recommended by easy seo. Some of them were blog posts, some of them were specific pages that target specific audiences with very specific calls to actions. For example, there’s a page for remote workers and there’s a page for families traveling to Breckenridge…etc. Those pages will capture different audiences with content that’s good for them. I loved it, and I would’ve spent hours contemplating on what to write. AI helped me in minutes!

Using our own tools, I got more traffic to the site and more interest. That was the immediate value that I was looking for. The fact that over the next six months I’ll get more traffic is nice, I think…

You see what I mean? The improvements to my website that I can see right now, that’s what would make me share EasySEO with friends, not the long-tail wins.

I may be right or may be wrong, but that's my theory right now. For the technical folks in the audience — we get so caught on in the details of what our superpowerful awesome tools can do, we forget the simple relevance to the end user.

Well, good news, this is an easy fix.

— Kirill.

Last week, I read about a startup founder who had just secured an investment. The story behind his check was revealing: the investor was impressed that he’d managed to cobble together significant functionality using no-code solutions. The investor’s logic was straightforward—if this founder could accomplish so much with limited tools, imagine what they could achieve with proper resources and a technical team.

It was impressive in its own way, like watching someone build a functional prototype with 100 PowerPoint slides. But it got me thinking about the work I create and how little I talk about it, not just with the outside world, but even with my own friends and cofounder.

Yes, I write this newsletter once in a while with some curious bit, but is that enough?

Take EasySEO, the product I’ve been building. On the surface, it does essentially the same thing today as it did two months ago — it analyzes your websites and tells you how to massively improve its SEO. But describing the iterations between then and now feels almost impossible. It’s like comparing a flip phone to an iPhone—both make calls and send texts, but the underlying complexity and user experience are night and day.

I spent many, many hours making the product what it is today.

The sheer volume of effort is staggering. Even my co-founder, who sees the git commits rolling in, doesn’t fully grasp the technical depth of what’s been built. I’m not saying this to brag—I genuinely question whether I should care about making this work more visible.

When someone does something non-technical, like stringing together 100 PowerPoints, non-technical people can immediately relate. “Wow, that’s a lot of work!” they say. But when you’re doing deep technical work? To most people, it’s just gibberish. It doesn’t look impressive because they can’t parse what they’re seeing.

If building with no-code tools is like making 100 PowerPoints, what I’ve been doing is like making 10,000—except the people who appreciate PowerPoints wouldn’t even begin to understand the medium I’m working in.

Here’s the kicker. I don’t think it’s all that impressive either. Only the end result matters, not the work, or the sacrifice, or the technical challenges.

I had a conversation with another founder friend this morning who’s building an LLM-powered operating system. It could be revolutionary, but I guarantee no one outside a small circle has any idea what he’s actually building. He’s the type who doesn’t care to explain it to people who don’t care to understand.

This creates a peculiar gap in our attention economy. Simple, easily understood projects get celebrated as if they’re incredibly complex. In five years, 99% of these will crash and burn, forgotten because they weren’t that complicated after all. But right now? They command the spotlight.

I like to think of myself and my technical founder friends as following the Elon Musk playbook. In the early days, nobody cares because what you’re building looks like a toy. The hope is to reach the point where your “toy” becomes threatening enough to established players that it transforms from something dismissible into something worth attacking.

Maybe that’s the real measure of technical work—not whether it impresses people today, but whether it survives long enough to become impossible to ignore. The PowerPoint builders get the early attention, but the deep technical work? That’s what actually changes the world.

---

Kirill.

p.s. So, should we care about making our technical work more visible, or should we keep our heads down and build until the work speaks for itself? I’m still figuring that out.*

p.p.s. How I feel lately, on one hand. On the other hand, if this were true, a donkey would be the kind of the animal kingdom.

It started simple. I wanted to boost my reach without shelling out for ads or whatever influencer/meme nonsense is hot this week. My theory was by learning about my follower’s interest, and the interests of their connections, I could drag their connection into the chatter, and create a ripple affect. This would enable me to keep saying what I want, but aim it at the people who are most likely to engage.

Could this be done? In theory, all of this data existed in the X api, but in practice, it wasn’t so simple. X API is a bit expensive, it costs $42,000 a month for unlimited data, real-time everything. $5,000 gets you “decent” access, but simultaneously too much for a side project, and too little to get it all. I ended up shilling out $200 for the basic crumbs.

Querying X against a basic API isn’t impossible, but it requires getting creative. The biggest challenge is API limits on speed and number of records you can get, plus a very stiff restriction of what you can actually get. Token limits mean you gotta batch it, nurse it along, and hope that you got what you needed, before trying again, all while your download speed is like 1999.

You can see in the screenshot below that I was able to get some information on who I follow, who follows me, what kind of interactions we have between each other, and I am also able to tap into their followers on demand.

The key here - you can’t look into the past, but you can run this tool every day to keep updating the information. Here’s where it gets really interesting —

Imagine you’re not scraping pennies—you’re the NSA, CIA, Anduril, Palantir at el. Your budget it a black box of infinite possibilities. You can farm so much information!

Spin up a dozen accounts, each with a distinct vibe. No need for fancy AI puppet. Actually hand those accounts to sharp folks with agendas and a field guide on how to steer controversy. Create accounts that focus on specific subjects, and then talk about that subject, arguing for both sides at the same time.

One account hates Trump, the other one loves him, one account hates war, the other wants to fight, one is advocating crypto growth, the other is predicting a crash at every turn…etc. Do this on a massive scale and tap into the data.

Every like, retweet, response? It’s a data point. Polar opposites mean you triangulate: Who bites on the lefty bait but ghosts the crypto grind, who loves the pew pew, who is likely to show up at a protest …etc?

Over months, you’ve got psych profiles for voter swings, brand loyalties, likelihood to act in one way or another… You can get a very, very detailed picture of specific targets, or a very broad and accurate understand of groups. If you want to segment those groups, divide them them further using a rhetoric from your other accounts and see how the groups split.

I’m running this on the cheap, trying to squeeze the most out of the little scraps of data, and I would love you to join me. $100 / month and you can start asking questions about your followers. I would love to get 10 new users to start asking more questions, and eventually get to 100 more users, so we could tap into the next tier of the API access.

The more each one of us learns from this system, the more we learn together, and visa-versa.

Got something you are dying to know about your follower graph?
Email me, and I can onboard you into the product.

— Kirill.

Great. So here’s what happened. I was driving from Austin to Breckenridge and I got tired of audio books and started chatting with Grok. The AI enlightened me about a ton of interesting new developments in material science, from self-healing concrete to cloud-like materials that crystalize into razor sharp glass. What I really wanted to know was how to teach all of this stuff to my kids, and that’s when we got to NiTi.

Nitinol is a mind-bending nickel-titanium alloy that “remembers” its shape and then snaps back into it, when heated. It’s the stuff behind medical stents that expand in your arteries, certain car components, and robot muscles.

Cool, I thought. We can definitely demo that at home. I got the wire from Amazon, and gleefully demonstrated how a spring uncoils itself into a straight line when submerged into hot water.

Kids were happy. There were some wows, but no applause. Kids still wanted to know why, why does this happen? Talking atoms is one thing, but seeing them?
AI to the rescue!

I spent an hour at a coffee shop talking to Claude Code about what I wanted to do, and the tool was born.

On the left: The high-temp austenite state, a perfect grid of yellow spheres (nickel and titanium atoms) marching in symmetric rows.
On the right: The low-temp martensite state, all sheared and staggered, gray and yellow atoms sliding past each other in a chaotic tangle.

Now, when I show my kids the Nitinol wire experiment (twist it cold, heat it with a hairdryer, watch the recoil), I can pull up this and say, “See? The atoms on the right want to regroup, they want to be in the symmetrical composition and don’t want to be bent. Heat just gives them the nudge.”

Okay, fine, they probably still don’t understand. But…

It will take more effort on my part, but I can keep expanding the tool outwards, to be able to zoom in and out between a write, and its atomic structure, eventually visualizing the bend, and then snap back. It’s more work, but it’s not impossible.

We don’t need to be waiting for AGI to solve “everything,” we just need to use whatever AI we got to solve the problems in front of us, turning “what if?” into “watch this!”

I am hoping for a future when we have more time to solve these problems in front of us. What do you say—ready to bend reality?

— Kirill

p.s. If you want more science, read my transcript discussion with Grok. Well, the short version, I will spare you the 15 hour drive one.

p.p.s. Nitinol Phase Visualizer is now on Github.

I began to realize this when so many VCs turned to support the Biden administration, which was notoriously anti-business. The 2+2 just didn’t make sense to me. Then I realized —

VC is system where public money is pooled, controlled by a select few, and doled out to insiders who cruise through life on handouts, regardless of success or failure. This isn’t the wild west of entrepreneurship; it’s a centrally planned economy run by elite bureaucrats who skim off the top while the real producers grind away. Socialism at its very best!

Is everyone in the VC world guilty of it? No, I think there are some people who genuinely believe they can make a different and to guide us towards a better, more advanced, more prosperous future. VC is just a mechanism to get there.

How many think that way vs just want to hitch a ride on the gravy train?
Hard to say.

Public Money, Private Control

At its core, venture capitalism starts with raising massive funds—not from scrappy individual investors, but from vast pools of institutional capital. Sure, occasionally you’ll hear about exited founders chipping in, but on the grand scale—think $100+ million funds—the money comes from pensions, university endowments, 401(k)s, and other “public” sources. These are essentially taxpayer dollars, funneled indirectly through everyday people’s retirement savings.

VCs position themselves as superior investors, claiming they can spot winners better than the market. But what they’re really doing is acting like central planners in a socialist state: taking collective resources and deciding who gets what.

In true capitalism, you’d bootstrap or compete for loans based on proven merit. Here? It’s a handout from the masses to the chosen few. And let’s not forget—the VCs get paid upfront. A typical 2% management fee on that pooled capital means they’re compensated just for holding and distributing public money, much like socialist bureaucrats who earn salaries for overseeing state resources without ever producing a thing.

A Controlled Economy with Insider Handouts

Once the money’s in, VCs distribute it selectively to startups they deem worthy. Sounds efficient? Hardly. Most ventures fail spectacularly, but that’s not the point.

The real game is control. Acquisitions often aren’t organic market outcomes—they’re VCs shuffling their portfolio like a deck of cards, merging one failing bet into another to keep the illusion alive.

I’ve known founders who’ve tanked two, three companies, yet live comfortably. Why? They’re in the circle. As long as you’re part of the club, you get another shot, another infusion of cash. It’s a safety net of mutual back-scratching: everyone participates in giving each other a share. Just make sure to say only the good things about your handlers.

This is textbook socialism—handouts for the connected, regardless of output. Contrast that with true capitalism, where small business owners hustle nonstop without injections or networks to fall back on. No one’s bailing you out if your revenue dips; you work your ass off or fold.

Rules for thee, but no for me

The VC ecosystem sells itself as a brutal fight to the top, where only the fittest survive. But insiders? They cruise. Small businesses are the ones truly fighting—always one bad quarter from disaster, with no safety net. Even if you’re profitable today, tomorrow’s uncertain because banks enforce strict rules: no loans without collateral and proven cash flow.

Remember the Silicon Valley Bank (SVB) meltdown? The freakout was confined to VCs and startups because they’d grown addicted to sweet deals.

Zero-revenue companies raising hundreds of millions could snag tens of millions in credit lines, all while depositing back into SVB in a cozy circle of mutual enrichment. Meanwhile, a small business turning a million in profit? Good luck getting a matching loan—banking regulations shut them out. VCs bypass it all through their networks, proving it’s not merit, but connections that win.

And when SVB wobbled? The government, ushered by your favorite deal makers, quietly stepped in to prop it up, echoing how socialist regimes subsidize failing state enterprises.

It was laughable how many VCs at the time were talking about “real jobs and real businesses” that allegedly relied on SVB, with ZERO real consequences. Meanwhile, actual businesses, with actual cash in profits, were struggling to raise a bare minimum to grow their operations.

Bureaucrats Skimming from Producers

This isn’t hyperbole; the parallels to historical socialism are stark. In the Soviet Union party bureaucrats got cushy perks (salaries, dachas, privileges) for doling out collective resources from farms and factories, without ever touching the tools. Workers produced; elites skimmed.

VCs do the same: 2% fees upfront, plus 20% carried interest on exits. If the startup flops? They keep the fee, blame the founders. If the start is a success, they attribute it all to their own genius at picking the winner.

Statistically, VCs are like mutual funds - on average you get better returns by sticking with S&P 500.

Look at flops like WeWork or Theranos—insiders walked away unscathed, no clawbacks. It’s subsidized failure, just like Soviet tractor factories propped up despite global irrelevance. Private farmers starved while the system funneled resources to the inefficient elite. In VC, your 401(k) funds the yacht parties, while small biz owners crawl.

Anti-Business Policies

This whole realization hit me during the Biden era. Why were so many VCs—self-proclaimed stewards of capitalism—adamantly supporting an administration seen as anti-business, with regulations and taxes galore? Meanwhile, others defected to Trump.

Two reasons emerged. First, sunk costs in relationships. If you’ve spent years building ties with Democrats in Congress, the Senate, and the White House, switching sides torches bridges. It’s expensive and risky, so you double down, hoping for future favors. This is pure favoritism: advancement based on who you know, not what you produce—another socialist hallmark, where connections trump merit. I understand it though, it makes sense.

Second, some VCs are outright socialists at heart. They milk the system, taking public money, charging fees, and living comfortably in the club. Especially if they came from money and had never worked a day in their lives, it makes perfect sense that someone else works, while they vacation. Biden’s world of handouts and regulations suits them fine.

But the Trump supporters? They’re the outliers who believe in duty—investing not just for today’s buck, but for a prosperous future. They want tech that scales, innovations that endure, because they’ve got enough money and status. It’s still riding the system, but with a nod to long-term societal gain over personal hoarding.

Is it a coincidence that it was entrepreneurs that came from nothing, like Elon Musk and Mark Andreessen, who were among the folks to swing over to the Republican camp?

No Consequences, No Change

In this socialistic VC machine of self-dealing, no one’s going to jail, going broke, or facing real trouble. If you’re in the club, life’s golden—endless second chances, bailouts, and comfort. Even if you crush and burn through a fund, or a few, it will still take a few decades to materialize, and someone else will pick you up into their firm then. You will safe face, and continue. But for the uninvited? Small business owners struggle eternally, future unsecured, no free money in sight. We’re the ones funding their game through our pensions, while they pretend it’s capitalism.

Is there an alternative system? Bootstrap is the only one I can think of. Make it work, and then just carry on that way; less glamor, but also a more free mind and soul.

Socialism for the rich, capitalism for the poor, that’s just not something I care about.

-Kirill.

This isn’t an accurate method nor does it work all the time, but it seems to me like the best way to weed out the liars is to find the people who don’t have a filter.

There’s a very good saying that if you don’t ever lie, you don’t have to remember what you said. People that lie have to keep a very good mental tally of what they think, vs. what they are tying to pretend to think in pursuit of a narrative. Personally I am very cautious of people that are always positive. I think they are mostly full of shit.

No one is happy and positive 100% of the time unless they are trying to hide something very deep inside, and if they are able to hide something this well, you’ve got to be careful - you can never trust that what comes out of their mouth is truth. If they lie to someone else, then why not you?

On the other hand, people that swear and have no filter, people that just tell it like it is, they are less likely to fool you. They might still have a greater goal, they might still talk with a narrative in mind, but at the very least they will be the type of people to punch you in the face, instead of stabbing you in the dark, in the back.

That’s it. That’s all i’ve got for today.
If you are going to pick someone to trust, pick a straight shooter.
It might not always work out, but your adds are better this way.

-Kirill.

Generally, when people tell you to focus, they mean that you should work on just one thing because that ought to be your obsession, and the thing you spend all the time on.

They generally don’t recommend that you work on 10 different things in parallel, while also trying to read, hike, spend time with family, exercise, eat health …etc etc etc.

Is that a good advice? yes or no.

I've been thinking about this a lot lately, and the way I've come to understand focus is this.

Focus is the ability to assign resources to projects in a way that improves the odds of a positive outcome.

Take Elon Musk, for example. Do you think he is not focused? He's working on multiple companies at the same time, and all companies are doing pretty well. Same goes for Jack Dorsey a few years ago, he was working multiple companies, and the companies did well. By that accord you can multitask just fine and you don't need to focus as long as you get shit done.

The key is that these guys didn’t do it alone; they had sufficient resources to hire teams that did the work. They raised the money and the pushed the boundaries, but really, they just managed resources better than anyone else.

Most people giving you advice to only work on one thing, simply assume linear distribution of resources and they assume that you only have enough time and energy and mental capacity to do just one thing. That may be true. However, that does not have to be true. You can make it not true.

The reality is that most things don't demand 12-18 hours of consistent effort per day, so if you are able to effectively context switch between five different tasks in that time spam, and you can provide the maximum output on those five things, then you absolutely can and should multitask.

Most people are gonna spend eight hours a day doing one thing poorly and you have an option to do five things well; do that and don't listen to anybody's advice on what it means to focus because only you know what that means, and how far you can push it.

Most people don’t run on jet fuel, you might.

You do you.

-Kirill.